Parallel Computing

coleslawokraSoftware and s/w Development

Dec 1, 2013 (3 years and 10 months ago)

60 views

Parallel Computing
Prof.Volker Strumpen
Institute for Computer Architecture
Johannes Kepler University
June 6,2013
Prof.Volker Strumpen Parallel Computing 1/8
Moore's Law
The number of transistors on a chip doubles every 2 years.
Transistor count of Intel microprocessors at date of introduction:
1000
10000
100000
1e+06
1e+07
1e+08
1e+09
1e+10
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Transistor Count
Year
Source:Wikipedia
Prof.Volker Strumpen Parallel Computing 2/8
The End of Frequency Scaling
Frequency scaling has reached limit due to power constraints.
Clock frequency of Intel microprocessors at date of introduction:
100000
1e+06
1e+07
1e+08
1e+09
1e+10
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Processor Clock Frequency
Year
Source:Intel
Prof.Volker Strumpen Parallel Computing 3/8
The Power Limit
The power limit is the new constraint (upper bound) in computing.
Power density of Intel microprocessors at date of introduction:
1
10
100
1000
10000
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Power Density [Watt / cm^2]
Year
hot plate
nuclear reactor
rocket nozzle
sun's surface
Source:Intel
hypothetical
VLSI limit
Prof.Volker Strumpen Parallel Computing 4/8
Industry Response
Stop frequency scaling and double number of cores each generation.
Die shot of IBM's 8-core Power7 chip multiprocessor (multicore):
Source:IBM,2009
Prof.Volker Strumpen Parallel Computing 5/8
State of the Art
64 cores in Dell PowerEdge R815 with 4 AMD Opteron 6276
processors,each with two chips of 8 Bulldozer cores.
Dell R815,2012
Prof.Volker Strumpen Parallel Computing 6/8
Cutting Edge Silicon
Intel's Xeon Phi integrates 62 cores on a single chip.
Intel Larrabee,2013
Prof.Volker Strumpen Parallel Computing 7/8
Learning Goals
Goal:Master the design principles behind parallel programming for
multicores with the multithreaded programming language Cilk.
Theoretical and algorithmic foundations of parallelism
Basic parallel algorithms:multidimensional parallel loops,map
and reduce,prex computations,pointer jumping,list
contraction,treex computations
Advanced parallelization techniques for matrix computations,
graph algorithms,search problems,and others
Algorithmic techniques for high-performance programs,such
as cache oblivious algorithms
Energy complexity of parallel computations
Prof.Volker Strumpen Parallel Computing 8/8